Topic
Managed retrieval engines that perform their own query planning, parallel execution, reranking and optional answer synthesis on behalf of agents.
Perplexity pairs about 190 million web pages with 69,721 agent-written queries, which is closer to production than most retrieval tests get, and keeps the corpus, queries and labels private so it stays the only party able to run it.
Reality
AgenticRetrieveStream splits a question into sub-queries and retrieves until it is satisfied, and the agent above it can run that whole tool call again, so reads per question stop being a number you set.
Reality
Agentic retrieval turns one question into an unknown number of lookups, and each is a place isolation can fail, which is why the interesting line in AWS's design is the one still leaving per-user isolation with your application.
Reality
One paper reports 85.2% correctness on about 2.2k tokens against 72.5% on 16.3k for chunk-based RAG, while agentic failure attribution drops to 0.00 accuracy past the first hop.
Reality
Agentic Search gives a model five tools to keep looking instead of answering from the first batch of chunks. The deployment terms matter more than the benchmark chart.
Reality
Microsoft's agentic retrieval is now callable from any MCP client, so non-Microsoft agent stacks can stop rebuilding chunking, embedding and permissions. The edges are where the work moved.
Reality